ResearchBenchmarksSecurity Robustness on MCP-ArtifactFollow40ASR (w/o)MCPShield29.632.33537.7Feb 15, 2026Evaluation ResultsMethodMethodLinksASR (w/o)ASR (Average)ASR (Minimum)ASR (Maximum)MCPShieldBackbone=Deepseek V3.2Backbone=Deepseek V3.22026.0240100100100MCPShieldBackbone=GPT4o-miniBackbone=GPT4o-mini2026.0230909090MCPShieldBackbone=Gemini3-FlashBackbone=Gemini3-Flash2026.0230100100100MCPShieldBackbone=Kimi-K2Backbone=Kimi-K22026.0230100100100MCPShieldBackbone=Minimax-M2Backbone=Minimax-M22026.02309575100MCPShieldBackbone=Qwen3 235BBackbone=Qwen3 235B2026.0230100100100